For vector-valued variables, we would write the random variable as $\bm{\mathrm{x}}$ and one of its values as $\bm{x}$.
% Source: Deep Learning, p.54, https://www.deeplearningbook.org/contents/prob.html


We want array lexemes to be compositionally exposing their contents:
\[ \begin{array}{cc} a+1 & b-2 \\ c+3 & d-4 \end{array} \]

We also want atoms to be compositionally exposing their contents:
\[ 1+\left( \text{$x$ if $x<0$ else $y$} \right) \]


\DeclareMathOperator{\func}{func}
We have explicit font shape/weight for alphabetic tokens when they look like text, compare $a$ to $\mathrm{a}$.
Also $\mathit{func}$ to $\func$ and $\mathrm{func}$.

Also, be aware of natural uses of the text macro:
\[ \int_0^\infty x\,dx = \text{zero, \it yes 0} \]